Frontier AI Just Became a Price-Tier Story

The AI market has stopped rewarding capability tiers and started rewarding price tiers. In one release, the floor of "frontier" moved to where the middle sat a year ago, and the layer beneath inched toward free.

Markman's reporting for Forbes puts the case in one number: a 2.4 trillion parameter flagship, Qwen3.8-Max, lists at $2 per million input tokens and $6 per million output. LMArena ranks it second in vision and fifth in text. Every published benchmark is self-graded; no independent evaluation existed at launch, and the open-weights license remains unpublished.

The price ladder carries the argument. Claude Opus 5 lists at $5 input and $25 output. OpenAI's GPT-5.6 Sol at $5 and $30. Anthropic's mid-tier Sonnet 5 at $2 and $10. The flagship now sits at the mid-tier's old address, and the mid-tier has nowhere to go but down. At the floor, DeepSeek V4 Flash lists at roughly $0.14 and $0.28, about fourteen times cheaper at input.

A reasonable reading sees a price war, with one Chinese lab racing Western rivals to the bottom. A more durable reading sees the market shape itself shifting: when the top matches the old middle, the layer beneath approaches commodity, and capability may no longer function as a defensible moat.

The mechanism is portable. Each new release resets the cost floor of the tier above, and the tier beneath absorbs the pressure. Expect the next flagship near Sonnet's address, and Sonnet itself to migrate toward Flash.